Image Processing and Deep Learning Based Failure Detection for Maintenance Optimization in Photovoltaic Panels

Inspection in photovoltaic generation systems is of great importance for maintenance and corresponding generation efficiency. The combined use of unmanned aerial vehicles and deep learning computational algorithms can be considered a potentially strong technique for the automatic inspection of non-conformities present in photovoltaic panels. In this work, a methodology was developed for inspecting solar power plants by capturing videos collected by drones, processing them using a new segmentation algorithm, and classifying the resulting images by convolutional neural network models. Subimages cropped from the extracted images were processed and classified into one of four classes: intact panels and panels with mild, medium, and severe non-conformities. Accuracies of up to 98.7% in the classification of non-conformities were obtained.